Title: An intelligent photovoltaic fire risk prediction platform using hybrid models based on AI, sensors, and anomaly analysis
Project ID: FENG.01.01-IP.01-002/25
Project Leader: Fabryka Bezpieczeństwa FABE sp. z o.o
Partners: Lodz University of Technology
Implementation Period: 01.06.2026 - 31.05.2029
Description: The dynamic growth in the number of PV/PV+BESS installations (installed PV capacity [GW]: 2020 – Poland ~4; global ~766; 2024 – Poland ~21; global ~1,865) increases exposure to fires resulting from component degradation, electric arcs, and overheating, while current safety measures are primarily reactive and do not provide reliable prediction. This is reflected in the estimated annual value of losses from PV fires: the "average" cost in Poland [PLN million] – 2020 ~128, 2024 ~700; the "average" cost globally [USD billion] – 2020 ~6, 2024 ~16 (the above figures are likely underestimated due to inconsistent reporting systems).
Consequently, the lack of early warning tools translates into property losses, risks to users, and higher maintenance costs. To solve this problem and significantly increase the operational safety of installations, a tool for early anomaly detection (long before escalation into a fire) and forecasting critical events is required.
Our project focuses on the development and demonstration of an early warning platform for PV installations (including PV with battery energy storage systems), combining multi-channel sensing (fiber optic FBG/DTS sensors, electrical paths, and electromagnetic field measurements) with hybrid AI models that incorporate physical constraints. The goal is to achieve a prototype at TRL VI (Technology Readiness Level 6), ready for pilot deployment in an operational environment.
The scope of project work includes:
- Thermal characterization of PV installation components and defects, and the identification of heating patterns leading to critical events;
- Implementation and integration of FBG/DTS sensors with PV components, along with methods for environmental impact compensation;
- Development of a diagnostic path based on electromagnetic field measurements (near-field probes/antennas, spectral and temporal analysis of arc signatures, leakage currents, and EMC disturbances) and the fusion of this data with thermal and electrical channels;
- Data acquisition and building signature libraries, as well as "SCE cards" (Surrogate Critical Events) with prediction horizons;
- Validation of signal fusion methods and the "data → forecast → decision" decision-making engine in SIL/HIL (Software/Hardware-in-the-Loop) environments;
- Long-term testing in a field laboratory (field-lab), concluding with a prototype demonstration in a relevant environment.